Permission model
Also known as: agent permissions · authorization model
A permission model defines what an agent is allowed to do: which tools it can call, which data it can read or write, under what conditions.
- Read vs. write vs. execute permissions.
- Per-tool, per-directory, per-domain scoping.
- Enforced by the harness, not the model.
The key insight: the model *requests* actions; the permission model *decides* which are allowed. That separation is what keeps a capable model safely bounded.